Search papers, labs, and topics across Lattice.
3
0
3
LeanGuard achieves 82.90 F1 score with 100x less compute than traditional reasoning guards, challenging the need for complex reasoning in safety moderation.
Overconfidence in VLMs can lead to misleading responses, but Semantic Flip effectively trains agents to refuse unanswerable queries with high accuracy.
Achieving closed-source model performance with open-source methods, BinTrack redefines spatial question answering for service robots navigating complex environments.
